c/9787: optimization bug with 176.gcc from SPECcpu2000

c/9787: optimization bug with 176.gcc from SPECcpu2000

[thomas.anders]

>Number:         9787
>Category:       c
>Synopsis:       optimization bug with 176.gcc from SPECcpu2000
>Confidential:   no
>Severity:       critical
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Fri Feb 21 14:16:00 UTC 2003
>Closed-Date:
>Last-Modified:
>Originator:     Thomas Anders
>Release:        gcc version 3.2.2
>Organization:
>Environment:
Intel Pentium 4 1.70GHz
SuSE Linux 8.1, Kernel 2.4.19
>Description:
The 176.gcc benchmark from the latest SPECcpu2000 V1.2 benchmark suite (www.spec.org/osg/cpu2000/) gives wrong results ("miscompare") when using optimization levels 1, 2 or 3 (e.g. "-O3") and "-march=pentium4" with gcc 3.2.2 (Release). Unfortunately I can't provide the source code here since it's covered under SPEC's license. A SPEC licensee (from SuSE, RedHat and friends) should be looking at this issue.
>How-To-Repeat:
Compile the 176.gcc benchmark from the SPECcpu2000 benchmark suite (www.spec.org/osg/cpu2000/) with OPTIMIZE="-O3 -march=pentium4" using gcc 3.2.2 (Release).
>Fix:
Yes, please! ;)
>Release-Note:
>Audit-Trail:
>Unformatted:

Re: c/9787: optimization bug with 176.gcc from SPECcpu2000

[Janis Johnson]

The following reply was made to PR c/9787; it has been noted by GNATS.

From: Janis Johnson <janis187@us.ibm.com>
To: thomas.anders@blue-cable.de
Cc: gcc-gnats@gcc.gnu.org
Subject: Re: c/9787: optimization bug with 176.gcc from SPECcpu2000
Date: Mon, 24 Feb 2003 16:22:48 -0800

 > The 176.gcc benchmark from the latest SPECcpu2000 V1.2 benchmark suite (www.spec.org/osg/cpu2000/) gives wrong results ("miscompare") when using optimization levels 1, 2 or 3 (e.g. "-O3") and "-march=pentium4" with gcc 3.2.2 (Release). Unfortunately I can't provide the source code here since it's covered under SPEC's license. A SPEC licensee (from SuSE, RedHat and friends) should be looking at this issue.
 
 The source code of the benchmark is GCC and it's covered by
 the GNU General Public License; SPEC can't change that.  The
 input that causes this behavior, however, is a different
 matter.  (This isn't meant to be legal advice.)